Subject: Re: tape drive instrumentation round 2
To: YAMAMOTO Takashi <email@example.com>
From: Brett Lymn <firstname.lastname@example.org>
Date: 03/06/2006 23:07:31
On Mon, Mar 06, 2006 at 09:39:31AM +0900, YAMAMOTO Takashi wrote:
> nfs is not a drive.
well... I know it's a bit of a reach but someitmes it is referred to
as such. No worse than being a "disk" :)
> what exactly needs "drivenames", and does it care about nfs?
Nothing needs drivenames right now but it may be useful to know what
drives are on the system, I am thinking of perhaps a tape drive that
was not attached/powered during boot but later gets powered up and the
scsi bus rescanned to attach it - it could be hard to tell if a tape
device is known about by the kernel without poking random device names
to see if something responds. The nfs was there just for completeness.
> it doesn't seem compilable, given drive_find's prototype and usage.
Hmmm how odd. It certainly compiled for me and the kernel +
iostat/vmstat worked as expected.
> to me, it seems weird to make disk_* (or drive_*) functions take io_stats.
> i think it's better to make disk_* functions call iostat functions internally.
> in this way, there is no need to change disk drivers much.
Well, all most of the disk_* took the disk structure for in the first place
was to update the stats that were embedded into the original struct
disk. The reason I made drive_* take a stats structure was to avoid
having to have all the things I wanted to collect i/o stats on using
struct disk. Perhaps what I should do is rename the drive_* to be
iostat_* and then just have disk_* be a wrapper for those calls which
would reduce the changes.